concord ai
…
← All guides

Setup

Add an MCP server for multi-agent coordination

One command registers a local coordination MCP server with Claude Code, Codex, Cursor, Gemini CLI and Grok Build in a repository.

The problem

Every agent client has its own MCP config format and location. Wiring one shared server into all of them by hand is tedious and easy to get wrong.

With Concord

  1. 01

    Install

    npm install -g @concord-ai/concord-mcp installs the concord CLI and the concord-mcp server.

  2. 02

    Run setup in the repo

    concord setup creates .concord/, registers the server in .mcp.json, .cursor/mcp.json, .gemini/settings.json, .grok/config.toml and ~/.codex/config.toml, and merges into existing config.

  3. 03

    Restart and check

    Restart your agent clients, then run concord adapters status to see which harnesses are reachable and concord status to see the workspace.

Result

Every agent in the repo talks to the same coordination server, set up in one step.

Common questions

Can I manage MCP config myself?
Yes. concord setup --no-mcp writes only the workspace and instructions. You can then register the server yourself, for example with claude mcp add concord -- concord-mcp.
Does it send my code anywhere?
No. State is local SQLite in .concord/. Concord sends anonymous product telemetry that never includes code, paths or task content, and CONCORD_TELEMETRY_DISABLED=1 turns it off.

Further reading

Give your agents shared work-state.

npm install -g @concord-ai/concord-mcp

View on GitHub →